How Small Businesses Can Actually Use AI

The genuinely useful applications of AI in a small business are unglamorous. Reading an incoming enquiry and pulling out what it is about. Drafting a reply for a person to check and send. Summarising a long thread so whoever picks it up next has the context. Sorting things into categories consistently. None of it looks like the demos, and all of it removes work somebody is doing by hand today.

The distinction that matters

Most of what a small business needs is not AI at all — it is automation. A reminder going out before every appointment does not require intelligence, it requires something to happen reliably. Reach for AI only where a step genuinely needs language understood or unstructured information interpreted. Using it where a rule would do adds cost, latency and a new way to be wrong.

Where it earns its keep

Turning messy input into structured data — a free-text enquiry into a name, a job type and an urgency. Drafting the first version of a repetitive reply. Summarising. Classifying. All of these share a shape: the output is checked by a person, or the cost of being occasionally wrong is low. That is the test for whether an AI step belongs in your process.

Where it does not

Quoting prices. Making promises about dates. Anything a customer will hold you to, and anything touching money or safety. It can prepare those — gather the history, draft the wording, flag the urgency — but a person should be the one who sends it. That is not caution for its own sake: it is the difference between a tool your team trusts and one they double-check, which costs more than doing it themselves.

A realistic first project

Take enquiries arriving through your website, your inbox and your phone, and have each one turned into a consistent record — who, what they want, how urgent, which channel — delivered to one place with a draft reply attached. It is small, it is checkable, and it attacks the stage where most small businesses actually lose money.

How to tell if it worked

Someone got hours back, and the business responds faster than it did. If neither is true, it does not matter how impressive the technology is — and you should be able to answer that question within a fortnight, not a quarter.